AppRank: Who’s the fairest iPhone app of all?
Back in October we shared some details about our iPhone app search “secret sauce”, including a brief introduction to AppRank, our social scoring algorithm for iPhone apps. Since then, we’ve been *very* busy cooking up some enhancements to our ranking methodology, adding new data sources and tweaking the math behind the calculation to further improve the results we deliver.
This post will introduce a few of the recent changes - and foreshadows another post we’ll put out in a few weeks with some even more exciting additions to our app discovery engine.
AppRank Today
If you’ve been following AppStoreHQ, you know that in addition to our app search engine we offer two different flavors of app discovery:
- Our “Hottest Apps on the Web” list scans the leading iPhone and tech blogs worldwide to discover new app reviews and mentions, providing a constantly-updated view into the most-talked-about apps among the leading “official” voices.
- In contrast, our “Hottest iPhone Apps on Twitter” list monitors broad popular sentiment, gulping down the Twitter firehose, discovering app-related tweets and turning them into a structured list of the most-Tweeted-about apps around the world.
Despite their distinct data sources, both lists take a similar approach to assigning a relative value to each iPhone app they include. Not only do we count the number of unique mentions for each app within a specified time period, we also assign a unique weight to each source, based on an objective and consistently-applied scoring model.
How Do We Weight Our Sources?
In the case of our Hottest Apps on the Web (or ‘Blog Hottest’ for short), we use two different third-party sources - Alexa and Compete - to assign a score to for relative popularity to each source (you can view our canonical blog list and the current Compete and Alexa scores for each included blog here).
In the case of our “Twitter Hottest” list, we borrowed some smart ideas from the good folks at the Hype Machine to assign a Twitter User Score to each unique Twitter account. Our tuning of the score led it in a slightly different direction than theirs, but it accomplishes the same goal of turning down Twitter spammers and rewarding / valuing the more authentically influential voices on that platform.
For each list - and in combination for our search results - we apply our weighting algorithm to each discovered mention, so the more trusted an authoritative sources (whether blogs or Twitter users) receive more weight than those that have not yet established themselves with their audience.
Time Matters Too
Once we’ve come up with a raw count, and then weighted it by the authority of the contributing sources, we also apply a time-based decay to the weighted score before calculating our rankings and injecting them into our search index. Why does time matter? Because different apps have been in the App Store for different amounts of time, older apps will tend to accumulate “value” (in terms of quality mentions) when compared to newer apps. To ensure that all apps are evaluated on a level playing field, we “decay” the value of older apps and mentions to normalize the results and present an accurate snapshot of app popularity *right now*.
What’s Next for App Rank?
We already think AppStoreHQ delivers the best and most relevant iPhone app discovery + search experience anywhere. But that doesn’t mean we aren’t always coming up with new ideas to make our results even better. Thanks to our growing base of registered users, we’ve been building up a *very* interesting dataset of personal recommendations based on saved preferences. We’re in the final stages of testing on a radical expansion of our app discovery toolset, adding a highly personalized vector for app discovery based on what we believe to be the largest dataset of individual app recommendations outside of iTunes itself.
Interested? Watch this space…
